Modeling and Predicting Trust Dynamics in Human–Robot Teaming: A Bayesian Inference Approach
DOI: 10.1007/s12369-020-00703-3
URL: https://doi.org/10.1007/s12369-020-00703-3
archive: archived pipeline: cataloged verified
Summary
Bayesian-inference model of human trust dynamics during repeated interaction with a robotic agent (Int J Soc Robotics 2021). Models trust as a Beta distribution updated from observed agent performance, encoding three empirical properties: trust persistence across moments, asymmetric weighting of negative experiences, and stabilization with repeated exposure. Validated on a published surveillance dataset of 39 participants supervising four drones; identifies three trust-dynamics archetypes (Bayesian decision maker, oscillator, disbeliever).
Key finding
A personalized Beta-distribution Bayesian model predicts moment-to-moment trust in autonomy with RMSE = 0.072, significantly outperforming prior trust-prediction methods, and reveals distinct individual trust-dynamics types relevant to adaptive automation design.
Methodology
Computational modeling: Beta-Bayesian trust update with parameters fit by Bayesian inference; reanalysis of an existing human-drone supervision dataset to compare against published trust models.
Sample size: N=39 participants (reanalyzed dataset)
Quality score: 5 / 5